nomis5 - 8:29 pm on Sep 20, 2010 (gmt 0)
75% of the articles on one of my "now sold" sites was between 3 and 7 pages long. I used a very roughly similar pagination system as Tedster suggests.
The reason for doing it site wide rather than a small % as a test is that (absolutely no proof) Google seemed to understand what the relationship was between the pages if it was done on a large scale, sitewise.
I did get sidetracked by the BBC, copying one of their pagination schemes. Same principal but the links were in very different positions on the page. It failed miserably. I converted the pages back to the original positioning and page views increased.
For that site I'm talking about, all this occurred over extended time scales, typically 3 to 6 months before I saw any results. But this may be because of the subject matter of the site.
I believe that correct pagination makes the best use of the content available where there are long articles. Possibly I'm ignorant, but the use of H1 and H2 never had any effect on my site and I did some experimentation. This also refers to long term rather than short term experimentation.